Introduction

Some of the VIMC modelling groups use child mortality in their models. These datasets are commonly presented as Infant Mortality Rate (IMR) for under-1 year olds, and Under-5 Mortality Rate (U5MR). A small number of groups also use the 28-day Neonatal Mortality Rate (NMR).

UNWPP provide IMR and U5MR for 1950-2100, but they do not provide NMR. IGME, (http://childmortality.org) provide all three, IMR, U5MR and NMR, but they only provide estimates, not predictions for the future; their 2017 dataset runs only as far as 2016. Furthermore, the first time-point for IGME’s estimates varies from country-to-country, and even for the three data types within the same country.

However, it can be well argued from observation that IGME’s estimates appear to model closer to reality than UNWPP’s; IGME appears to record genuine child mortality events, whereas UNWPP shows evidence of considerable smoothing. IGME also provide child mortality data for some countries that UNWPP do not; Marshall Islands and Tuvalu are noteworthy examples, since they are countries that modellers must provide estimates for.

Which source to use?

For VIMC purposes, the accuracy of data to real-life events is of secondary concern; our primary concern is consistency, both in the sense that we want all modellers to be using the same demographic data, and we want all the variables in that demographic data to be consistent with each other.

In discussions, UNWPP acknowledged that the IGME’s child mortality data is of a higher accuracy towards reality, and if time and funding were available, a merging of the two data sources in a consistent way would be ideal. This however is not something we should expect to happen soon, perhaps ever.

In the meantime, the use of IGME’s child mortality data, combined with all the other fields from UNWPP would be inconsistent, with the population data and life tables of UNWPP not being matched to IGME’s mortality rates, which differ from UNWPP’s as we’ll see. In any case, we would need to extend IGME’s time range back to 1950 and forward to 2100 to use the two source together.

VIMC therefore asks that for VIMC work, the groups should use WPP’s data for IMR and U5MR, with the understanding this is not the ideal choice, but the best compromise. We will then provide Neonatal Mortality data, (which UNWPP do not), in a simple way, that is consistent with UNWPP, but also is guided by the IGME data.

Method

Neonatal Mortality Rate (28-day) represents a subset of the 1-year Infant Mortality Rate. Therefore, provided that NMR is less than IMR (since the rates are measured in deaths per live births), the two will be consistent with each other, and will not cause any consistency issues with UNWPP, since population is only given at yearly timepoints.

We therefore define our hybrid NMR as a scaling of UNWPP’s IMR, multipled by IGME’s NMR/IMR fraction. Since IGME’s time-range usually starts after, and always finishes before UNWPP’s, we assume the nearest available NMR/IMR proportion for time-points that are out of IGME’s range.

Results

Below are the comparisons of UNWPP’s Under-5 and Infant Mortality Rates, which Montagu serves, compared to the data from IGME. The graphs show that while there appear to be methodological differences between the two datasets, they are broadly in agreement.

Note also that Kosovo (XK) is represented neither by UNWPP, nor IGME. Serbia is used as a substitute; see the separate documentation about Kosovo for further discussion.

Finally, we show Montagu’s hybrid neonatal mortality rate, compared to IGME’s. Where the red line is solid, we are able to use IGME data to calculate the NMR/IMR scaling; where the line is dotted, either or both of those fields is not available from IGME for that time point, so we are using the nearest available data.